Machine learning-driven development of food liposomes: phospholipid extraction, formulation design, preparation control, and stability prediction.

Liposomes are important food-grade delivery systems for hydrophilic and hydrophobic bioactive compounds, but their performance depends on complex interactions among phospholipid source, formulation composition, preparation conditions, and storage environment. These coupled variables make rational design and scalable production challenging. This review summarizes the application of machine learning (ML) across the food liposome value chain, including phospholipid extraction and identification, formulation design, preparation process control, bioactive compound screening, and stability prediction. Particular attention is given to how ML integrates lipid composition, payload properties, and processing parameters to predict key liposome quality attributes, including particle size, polydispersity index, zeta potential, and encapsulation efficiency, thereby guiding formulation optimization. When combined with microfluidic fabrication and automated experimentation, ML can further enhance preparation control and process optimization. Current evidence indicates that ML has strong potential to accelerate food liposome development. This potential arises from its ability to model nonlinear relationships among formulation, processing, and quality attributes and to extract useful information from complex molecular, analytical, imaging, and process datasets. However, current progress remains limited by small and heterogeneous datasets, inconsistent characterization protocols, limited model interpretability, and insufficient transferability across lipid sources, payloads, and food matrices. Future progress will depend on standardized datasets, interpretable models, and closer integration of ML with high-throughput experimentation and validation in real food systems.